Eye Movement Patterns as Robust Biomarkers for Schizophrenia Identification Using a Novel Data Transformation Approach.

Although eye movement abnormalities are documented in schizophrenia (SZ), their translation into objective diagnostic biomarkers remains limited. In this study, we propose a novel identification framework that integrates a Sparsity-Scoring Kernel Entropy Component Analysis (SSKECA) algorithm with a multidimensional eye movement feature set. A total of 40 patients with SZ and 50 healthy controls (HC) completed a free-viewing task involving 100 distinct semantic images. The proposed SSKECA algorithm optimizes multidimensional feature representations to capture latent eye movement patterns characteristic of SZ. The SSKECA-AdaBoost model achieved competitive performance, with an accuracy of 0.933 and an area under the receiver operating characteristic curve (AUC) of 0.960. Notably, when restricted to only 25 highly discriminative images, the SSKECA-XGBoost model achieved an accuracy of 0.922. Feature ablation analyses not only reproduced previously reported eye movement findings but also highlighted additional atypical patterns. Misclassification analyses revealed more pronounced eye movement deficits in incorrectly classified SZ patients. Overall, the proposed framework translates complex eye movement patterns into robust indicators for subject-level identification, offering a practical and efficient tool to support objective assessment in SZ.